Network optimization method and device

By determining the reliability of the AI ​​model and self-update and correcting errors, the problem of insufficient reliability of the AI ​​model when the network environment changes is solved, ensuring the adaptability and optimization effect of network configuration information.

CN119094346APending Publication Date: 2024-12-06HUAWEI TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310665683.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-06-06
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

When the existing AI model changes in the network environment, the logic of the inference engine is not adapted, resulting in its reliability being unable to be guaranteed and the complex and changing network environment cannot be effectively optimized.

Method used

By obtaining the target network data, determining the reliability of the AI ​​model, and issuing network configuration information in a reliable manner, the target network configuration information is determined using the secondary dictionary mapping and nearest neighbor distance method, and the model aging detection and error correction are carried out to achieve Self-update and real-time error correction of AI models.

Benefits of technology

Ensure that the issued network configuration information is applicable to the current network environment, improves the reliability and adaptability of the AI ​​model, and realizes real-time optimization of the network environment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119094346A_ABST
    Figure CN119094346A_ABST
Patent Text Reader

Abstract

The invention provides a network optimization method and equipment, and relates to the field of communication. The method is used for determining and issuing network configuration under the condition of ensuring that an AI model is reliable. The method comprises the following steps: acquiring target network data; determining target scene feature information based on an artificial intelligence AI model according to the target network data; determining target network configuration information according to the target scene feature information; determining the reliability of the AI model; and under the condition that the AI model is determined to be reliable according to the reliability, issuing target network configuration information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of communication technology, and in particular, to a network optimization method and device. Background Art

[0002] In order to maximize the performance of the network, it is necessary to continuously adjust and optimize the various network configurations. Manually adjusted static configurations can no longer meet the current complex and changing network environment, so optimizing network configurations through artificial intelligence (AI) models is an important optimization direction.

[0003] The existing AI model for optimizing network configuration usually first inputs offline network traffic feature data with feature labels into the AI ​​model for model training. After the AI ​​model is successfully trained, the network traffic data can be input into the AI ​​model. Based on the inference engine of the AI ​​model, the AI ​​model can infer the corresponding optimal network configuration based on the network traffic data and issue it.

[0004] In the current AI model network configuration solution, the AI ​​model is trained based on offline network traffic data. As the network environment changes, the AI ​​model's reasoning engine logic may not necessarily still be adapted to the current network environment, and the reliability of the AI ​​model cannot be guaranteed. Summary of the invention

[0005] The embodiments of the present application provide a network optimization method and device, which can ensure that the network configuration is determined and issued when the AI ​​model is reliable.

[0006] In a first aspect, an embodiment of the present application provides a network optimization method and device, which are applied to network equipment. The method includes: acquiring target network data; determining target scene feature information based on an artificial intelligence (AI) model according to the target network data; determining target network configuration information according to the target scene feature information; determining the reliability of the AI ​​model; and when it is determined that the AI ​​model is reliable according to the reliability, issuing the target network configuration information.

[0007] In this possible implementation, before sending the target network configuration information, it is determined that the AI ​​model is reliable, thereby ensuring that the sent target network configuration information is reliable, that is, suitable for the current network environment.

[0008] In a possible implementation manner of the first aspect, the method further includes: constructing the AI ​​model based on offline network data.

[0009] In a possible implementation of the first aspect, the above-mentioned construction of the AI ​​dictionary model includes: determining the sample set information of the AI ​​model based on offline network data, the sample set information including scene feature information, category identification and network configuration information, wherein the scene feature information includes the bandwidth, latency, number of packet losses and congestion status of the network data, the category label indicates the sample category of the scene feature information, and the optimal configuration set indicates the network configuration information corresponding to the sample category of the scene feature information.

[0010] In a possible implementation of the first aspect, the above-mentioned determining the target network configuration information according to the target scene characteristic information includes: determining the corresponding target sample set information in the AI ​​model according to the target sample characteristic information; determining the corresponding target category identifier according to the target sample set information; and determining the corresponding target network configuration information according to the target category identifier.

[0011] In this possible implementation, the mapping relationship between the target sample set information and the target category identifier, the mapping relationship between the target category identifier and the target network configuration information, and the AI ​​model determine the corresponding network configuration information through secondary dictionary mapping to achieve network configuration information determination.

[0012] In a possible implementation manner of the first aspect, the above-mentioned determining the target sample set information in the corresponding AI model according to the target sample feature information includes: determining the nearest neighbor distance between the target sample feature information and the AI ​​model, the nearest neighbor distance being the nearest neighbor distance between the target sample feature information and multiple sample set information of the AI ​​model; determining the target sample set information of the AI ​​model corresponding to the nearest neighbor distance.

[0013] In this possible implementation, target sample set information is determined by nearest neighbor distance based on target sample feature information, and a method for determining target sample set information is specifically provided, thereby enhancing the feasibility of the embodiments of the present application.

[0014] In a possible implementation of the first aspect, the above-mentioned determining the target network configuration information based on the sample characteristic information includes: determining k sample set information of the AI ​​model adjacent to the target sample characteristic information; determining k network configuration information corresponding to the k sample set information; and determining the target network configuration information based on the k network configuration information.

[0015] In this possible implementation, the target network configuration information is determined based on multiple network configuration information, thereby reducing the impact of a single model sample on the network configuration information output by the AI ​​model.

[0016] In a possible implementation manner of the first aspect, the above-mentioned determination of the reliability of the AI ​​model includes: determining the nearest neighbor distance between the target sample feature information and the AI ​​model; and determining the reliability of the AI ​​model based on the nearest neighbor distance.

[0017] In a possible implementation manner of the first aspect, the method further includes: determining an aging index of a model sample of the AI ​​model.

[0018] In a possible implementation of the first aspect, the above-mentioned determination of the aging parameters of the model samples of the AI ​​model includes: determining the total number of hits of the model samples of the AI ​​model and the number of hits within a period T; and determining the aging index of the model samples of the AI ​​model based on the total number of hits and the number of hits within the period T.

[0019] In a possible implementation manner of the first aspect, the method further includes: when the aging index does not meet the aging degree requirement of the AI ​​model, deleting the aged model samples of the AI ​​model.

[0020] This possible implementation method ensures that the samples of the AI ​​model are not aged by deleting the aged model samples of the AI ​​model, thereby realizing real-time error correction of the AI ​​model.

[0021] In a possible implementation manner of the first aspect, the method further includes: determining whether the AI ​​model needs error correction; and if the AI ​​model needs error correction, deleting model samples that need error correction in the AI ​​model.

[0022] In a possible implementation of the first aspect, the above-mentioned determination of whether the AI ​​model needs error correction includes: obtaining network feature change information; determining an error correction index based on the network feature change information; and determining whether the AI ​​model needs error correction based on the error correction index.

[0023] In this possible implementation, error correction processing is performed on the AI ​​model through error correction detection to achieve self-update and error correction of the AI ​​model.

[0024] In a possible implementation manner of the first aspect, the method further includes: when it is determined that the AI ​​model is unreliable according to the reliability, determining the target network configuration information through configuration exploration.

[0025] In a possible implementation manner of the first aspect, the above-mentioned determining the target network configuration information through configuration exploration includes: performing feedback network configuration exploration; determining that the configuration exploration has converged; and determining the target network configuration information.

[0026] In a possible implementation of the first aspect, before performing feedback network configuration exploration, the method further includes: obtaining network status information; determining whether the network status is stable based on the network status information; and performing feedback network configuration exploration includes: performing feedback network configuration exploration when it is determined that the network status is stable.

[0027] In a possible implementation manner of the first aspect, when the target network configuration information does not meet preset requirements, the method further includes: performing random exploration; and re-determining the target network configuration information.

[0028] In a possible implementation manner of the first aspect, the method further includes: determining a corresponding target sample category according to the target network configuration information;

[0029] In a possible implementation manner of the first aspect, the sample set information of the above-mentioned AI model does not include the target sample category, and the above-mentioned method also includes: adding a target sample category corresponding to the target network configuration information in the AI ​​model.

[0030] In a possible implementation of the first aspect, the method further includes: updating the sample set information of the AI ​​model according to the target network configuration information, the target sample category and the target scene feature information

[0031] In this possible implementation, network devices can determine the optimal configuration through online configuration exploration, and can also conduct random exploration based on the distance between the current exploration convergence status and the target state to prevent local optimality. This enables autonomous real-time online incremental learning of AI models without human intervention, ensuring good generalization of the model in unknown scenarios and ensuring the optimization effect of unknown scenarios.

[0032] In a second aspect, an embodiment of the present application provides a network device, including: a processor and a memory. The processor is coupled to the memory; the memory is used to store computer instructions, and the computer instructions are loaded and executed by the processor to enable the network device to implement any one of the methods provided in the first aspect.

[0033] In a third aspect, an embodiment of the present application provides a chip, comprising: a processor and an interface circuit; the interface circuit is used to receive code instructions and transmit them to the processor; the processor is used to run the code instructions to execute any one of the methods provided in the first aspect.

[0034] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which at least one computer program instruction is stored, and the computer program instruction is loaded and executed by a processor to implement any one of the methods provided in the first aspect above.

[0035] In a fifth aspect, an embodiment of the present application provides a computer program product, including computer execution instructions, which, when executed on a computer, enable the computer to execute any one of the methods provided in the first aspect.

[0036] The technical effects brought about by any implementation method in the second to fifth aspects can refer to the technical effects brought about by the corresponding implementation method in the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of a scenario for optimizing network configuration for an AI model;

[0038] Figure 2 A schematic diagram of the architecture of a network device provided in an embodiment of the present application;

[0039] Figure 3 A schematic diagram of a data center network scenario provided in an embodiment of the present application;

[0040] Figure 4 A schematic diagram of the architecture of a data center network scenario provided in an embodiment of the present application;

[0041] Figure 5 A schematic diagram of the architecture of another data center network scenario provided in an embodiment of the present application;

[0042] Figure 6 A schematic diagram of the architecture of another data center network scenario provided in an embodiment of the present application;

[0043] Figure 7 A flowchart of a network optimization method provided in an embodiment of the present application;

[0044] Figure 8 A flowchart of a configuration exploration method provided in an embodiment of the present application;

[0045] Fig. 9 A schematic diagram of a scenario of a configuration exploration method provided in an embodiment of the present application;

[0046] Fig.10 A flowchart of another configuration exploration method provided in an embodiment of the present application;

[0047] Fig.11 A schematic diagram of the structure of a network device provided in an embodiment of the present application;

[0048] Fig.12 A schematic diagram of the structure of another network device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] In the description of this application, unless otherwise specified, " / " indicates that the objects associated with each other are in an "or" relationship, for example, A / B can represent A or B; "and / or" in this application is merely a description of the association relationship between associated objects, indicating that three relationships may exist, for example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural.

[0050] In the description of this application, unless otherwise specified, "plurality" means two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0051] In addition, in order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, words such as "first" and "second" are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit the difference.

[0052] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way for easy understanding.

[0053] It is understood that the "embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the various embodiments in the entire specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It is understood that in various embodiments of the present application, the size of the sequence number of each process does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0054] It can be understood that some optional features in the embodiments of the present application may be implemented independently in certain scenarios without relying on other features, such as the solution on which they are currently based, to solve corresponding technical problems and achieve corresponding effects, or may be combined with other features according to needs in certain scenarios. Accordingly, the devices provided in the embodiments of the present application may also realize these features or functions accordingly, which will not be elaborated here.

[0055] In this application, unless otherwise specified, the same or similar parts between the various embodiments can refer to each other. In this application, if there is no special description or logical conflict between the various embodiments, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other, and different embodiments can be combined to form new embodiments according to their inherent logical relationships. The following implementation methods of this application do not constitute a limitation on the scope of protection of this application.

[0056] Network optimization is an important topic in the network field. In order to maximize the performance of the network, it is necessary to continuously optimize various network configurations. With the development of the times, the network environment has become increasingly complex. For example, for network traffic, the traffic model is complex and changes frequently. If manual optimization is used, it requires high experience from professionals and it is difficult to ensure the optimal service performance. On the other hand, manually tuned static configurations cannot adapt to the changing network environment and can only achieve partial scene optimization. Therefore, an automated network optimization method is needed to meet the requirements of network performance.

[0057] With the development of artificial intelligence (AI), the application of AI in the network field has become an important topic. For the optimization of network configuration, network configuration optimization based on data-driven AI models is an important research direction.

[0058] like Figure 1 As shown in the figure, the existing AI model for optimizing network configuration usually collects network traffic data offline first, then adds feature labels to the offline network traffic data, and then inputs the offline network traffic data with feature labels, i.e., network feature information, into the AI ​​model for model training. After the AI ​​model is successfully trained, network traffic data can be collected through a network traffic monitor, and then the network feature information can be determined through preprocessing. The network traffic feature data is input into the AI ​​model, and based on the inference engine of the AI ​​model, the AI ​​model can infer the corresponding inference result, i.e., the optimal network configuration, based on the network traffic data and issue it.

[0059] In the current AI model network configuration solution, the AI ​​model is trained based on offline network traffic data. As the network environment changes, the AI ​​model's reasoning engine logic may not necessarily still be adapted to the current network environment, and the reliability of the AI ​​model cannot be guaranteed.

[0060] Based on this, an embodiment of the present application provides a network optimization method, which includes: obtaining target network data; determining target scene feature information based on an artificial intelligence AI model according to the target network data; determining target network configuration information according to the target scene feature information; determining the reliability of the AI ​​model; and when the AI ​​model is determined to be reliable according to the reliability, issuing the target network configuration information.

[0061] like Figure 2 As shown, the network device applied by the network optimization method in the embodiment of the present application may include the following modules:

[0062] AI module 201: AI module 201 is the core module of the present invention, which is used to carry the AI ​​machine learning model of the embodiment of the present application, perform network optimization reasoning based on a two-level mapping dictionary, realize model iterative update based on an online incremental learning method, and perform model aging detection and error correction according to the real-time effect of the model.

[0063] The control module 202 mainly includes an aging unit 203 , an error correction unit 204 , an increment unit 205 and other units.

[0064] The functions of each unit of the control module 202 are as follows:

[0065] Aging unit 203: used to establish an AI model entry-inference table, periodically record AI model inference conditions, and perform aging detection on AI model entries based on model inference statistics.

[0066] Error correction unit 204: establishes a feedback mechanism based on the model output and implements model error correction based on changes in network characteristics.

[0067] Incremental unit 205: carries the online incremental learning mechanism of the AI ​​model, analyzes the reliability of the current AI model based on the reliability of the model, determines whether to add new sample categories based on category similarity, and constructs a two-level mapping sample for incremental learning of the AI ​​model.

[0068] Exploration module 206: used to carry the online exploration mechanism, perform network exploration based on the state machine, explore the optimal configuration, and provide input for the online incremental learning of the model.

[0069] Reasoning module 207: used to send network configuration information to the network according to the results output by the AI ​​module.

[0070] Collection and processing module 208: used to periodically monitor the network environment, collect network traffic information and perform pre-processing, and provide input for other modules.

[0071] In addition, the network device may also include other modules for implementing the network optimization method of the embodiment of the present application, which are not specifically limited here.

[0072] like Figure 3 As shown, the network optimization method provided in the embodiment of the present application can be applied to Figure 3 The data center network scenario shown in the figure, the network optimization method provided in the embodiment of the present application can be applied to Figure 3 In addition, the present invention can also be applied to other network scenarios, such as wide area network and campus network, which are not limited here.

[0073] In the embodiment of the present application, the network device to which the network optimization method is applied may include: Figure 2 As shown in the figure, all modules are in the same network device. In addition, these modules can also be distributed in different network devices in the same network, and multiple network devices can communicate with each other to transmit relevant information. The details are as follows:

[0074] like Figure 4 As shown, in a possible implementation, the data center network scenario includes a management switch, a top of rack (Tor) switch, a service node device, and a management node device. The service node device and the management node device can be a server, and the network optimal configuration issued by the present invention acts on a network interface card (NIC). Among them:

[0075] The server where the first NIC card is located and the server where the second NIC card is located are used as business node devices, on which exploration modules, reasoning modules and collection and processing modules are deployed. The business node device collects and preprocesses network scene feature information based on the collection and processing module, and sends it to the management node device for model reasoning. After receiving the reasoning result from the management node device, the reasoning module sends the optimal network configuration. The business node device can also perform exploration based on the exploration module.

[0076] The server where the third NIC card is located serves as a management node device, on which an AI module and a control module are deployed. The management node device can receive network scenario feature information sent by the business node device, output inference results based on the AI ​​model and transmit them to the business node device; the management node device collects and calculates information such as model reliability and model benefit based on the control module, performs aging detection and error correction on the model, and determines whether to start the incremental learning process and other mechanisms.

[0077] In an embodiment of the present application, the AI ​​module and the control module can also be deployed on a Tor switch, and each Tor switch is responsible for all business nodes under the Tor switch.

[0078] In the embodiment of the present application, the management node can be as follows Figure 4 As shown in the figure, under the Tor switch, in addition, data can also be collected and sent through the management plane under the management switch, and the specific details are not limited here.

[0079] In the embodiment of the present application, the network optimization method can be used to optimize the parameters of the NIC card. In addition, it can also be applied to the optimization of switch parameters. When optimizing the switch parameters, each module deployed in the service node device can be deployed on the Tor switch, and each module deployed in the management node device can be deployed on the management switch.

[0080] like Figure 5 As shown, in another possible implementation, the data center network scenario includes a management switch, a top of rack (Tor) switch, a service node device, and a management node device. The service node device and the management node device can be a server, and the network optimal configuration issued by the present invention can act on the NIC. Figure 4 The main difference between the data center network scenarios shown is the deployment location of each module, where:

[0081] The server where the first NIC card is located and the server where the second NIC card is located are used as business node devices, on which exploration modules, reasoning modules and acquisition processing modules are deployed. The business node device collects and preprocesses network scene feature information based on the acquisition processing module, and the AI ​​module performs model reasoning. After receiving the reasoning results from the AI ​​module, the reasoning module issues the optimal network configuration. The business node device can also perform exploration based on the exploration module.

[0082] The server where the third NIC card is located serves as a management node device, on which a control module is deployed. The management node device collects and calculates information such as model reliability and model benefit based on the control module, performs aging detection and error correction on the model, and determines whether to start the incremental learning process.

[0083] In an embodiment of the present application, the AI ​​module and the control module can also be deployed on a Tor switch, and each Tor switch is responsible for all business nodes under the Tor switch.

[0084] In the embodiment of the present application, the management node can be as follows Figure 5 As shown in the figure, under the Tor switch, in addition, data can also be collected and sent through the management plane under the management switch, and the specific details are not limited here.

[0085] In the embodiment of the present application, the network optimization method can be used to optimize the parameters of the NIC card. In addition, it can also be applied to the optimization of switch parameters. When optimizing the switch parameters, each module deployed in the service node device can be deployed on the Tor switch, and each module deployed in the management node device can be deployed on the management switch.

[0086] In this possible implementation, the AI ​​module is deployed on the service node device, reducing the latency of network configuration information reasoning.

[0087] like Figure 6 As shown, in another possible implementation, the data center network scenario includes a management switch, a top of rack (Tor) switch, and a service node device. The service node device can be a server where a network interface card (NIC) is located. Figure 4 and Figure 5 The main difference between the data center network scenarios shown is the deployment location of each module. The business node device includes all modules. The business node device can realize the functions of the aforementioned business node device and management node device, which will not be described in detail here.

[0088] In the embodiment of the present application, the network optimization method can be used to optimize the parameters of the NIC card. In addition, it can also be applied to the optimization of switch parameters. When optimizing the switch parameters, each module deployed in the service node device can be deployed on the Tor switch, and each module deployed in the management node device can be deployed on the management switch.

[0089] In this possible implementation method, a purely distributed deployment method is adopted, which has the lowest deployment difficulty and can be implemented through a single network device; on the other hand, the modules are integrated into the same network device, and the delay of information interaction between modules is also the lowest.

[0090] In the embodiment of the present application, the service node device and the management node device may be a server, such as Figures 4 to 6 As shown, the network optimization method provided by the embodiment of the present application can be deployed on servers such as service node devices and management node devices. In addition, it can also be deployed on the network interface card (NIC) of the server, which is not limited in the embodiment of the present application. It can be understood that in the embodiment of the present application, the execution subject can perform some or all of the steps in the embodiment of the present application, and these steps or operations are only examples. The embodiment of the present application can also perform other operations or deformations of various operations. In addition, each step can be performed in a different order presented in the embodiment of the present application, and it is possible not to perform all the operations in the embodiment of the present application.

[0091] It should be noted that the message names between the devices or the names of the parameters in the messages in the following embodiments of the present application are only examples, and other names may be used in the specific implementation, and the embodiments of the present application do not specifically limit this.

[0092] like Figure 7 As shown, Figure 6 Taking the example that all modules shown are in the same network device, an embodiment of the present application provides a network optimization method, which includes the following steps:

[0093] 701. Network equipment builds AI models.

[0094] The AI ​​module of the network device builds an AI dictionary model based on offline data.

[0095] Specifically, in the AI ​​dictionary model built by the network device based on offline data, X i is the i-th sample set information in the AI ​​dictionary model, X i ={F i ; Y i ;P i},in:

[0096] F i =(f i 1 , f i 2 , f i 3 , …, f i N ), F i is the scene feature information preprocessed by the collection and processing module of the network device, N is the feature dimension, and the collection and processing module of the network device can determine the scene feature information by preprocessing the collected network statistical information K through f(K).

[0097] In the embodiment of the present application, the network statistical information K may include information such as bandwidth, latency, number of packet losses, etc. In addition, it may also include information such as congestion status, which is not specifically limited here.

[0098] In the embodiment of the present application, the preprocessing means of the collection and processing module for the collected network statistical information K may include linear calculation and normalization, etc. In addition, it may also be a binarization processing means, which is not specifically limited here.

[0099] Y i is the category label of the i-th item of the model, which is used to indicate the category of the sample of the i-th item of the model. The category label can be a label of the type of serial number, name, etc., which is not limited here.

[0100] P iis the network configuration information corresponding to the i-th sample of the model. The network configuration information P i is the relevant information of the optimal network configuration corresponding to the i-th sample of the model, that is, in the network scenario indicated by the i-th sample of the model, the optimal network configuration information given by the model is P i .

[0101] In the embodiment of the present application, the scene feature information in the AI ​​dictionary model and the category label constitute a first-layer mapping relationship, and the category label and the network configuration information constitute a second-layer mapping relationship. Therefore, after determining the scene feature information, the corresponding optimal network configuration information can be determined according to the second-layer mapping AI dictionary model.

[0102] In step 701 of the embodiment of the present application, the network device can build a dictionary model based on offline data. In addition, the network device can also directly use an empty model to directly perform subsequent steps, that is, the network device can also not execute step 701, and execute subsequent steps based on the empty model. The specific details are not limited here.

[0103] 702. The network device obtains target network traffic data.

[0104] The network device obtains the target network traffic data of the network.

[0105] Specifically, the network device obtains target network traffic data through a collection and processing module, and the target network traffic data is data related to the network traffic of the network environment.

[0106] 703. The network device determines target scene feature information based on the AI ​​model.

[0107] After obtaining the target network traffic data, the network device can pre-process the target network traffic data to determine the target scenario feature information based on the artificial intelligence AI model. For example, the target network traffic data can be the network traffic time series data of switch A. The network device can process the network traffic time series data through linear calculation, normalization, binarization and other processing methods to obtain the scenario feature information of switch A, such as bandwidth, delay, and packet loss.

[0108] 704. The network device determines target network configuration information according to the target scene feature information.

[0109] The network equipment is deployed with a trained AI dictionary model. After determining the target scene feature information, the network equipment determines the target network configuration information based on the AI ​​dictionary model.

[0110] Specifically, based on the secondary mapping AI dictionary model, after determining the target scene feature information, the network device can determine the target category corresponding to the target scene feature information, and then determine the target network configuration information corresponding to the target category.

[0111] In a possible implementation, target scene feature information N=(N1, N2, N3, ..., N n ), n is the number of feature dimensions. In the AI ​​dictionary model, the i-th feature M i =(M i1 , M i2 , M i3 , …, M im ), m is the number of samples of the AI ​​dictionary model. The incremental unit of the network device can calculate the nearest neighbor distance D based on the Euclidean distance Dis m , That is, the nearest neighbor distance D m It is the minimum value of the m Euclidean distances Dis between the target scene feature information N and the m samples of the AI ​​dictionary model.

[0112] Then the network device calculates the nearest neighbor distance D m Determine the corresponding target sample, and based on the secondary mapping of the AI ​​dictionary model, first determine the target category, and then determine the target network configuration information corresponding to the target category.

[0113] In another possible implementation, the network device may not calculate the nearest neighbor distance D m , but determine the first k samples with the closest neighbor distance X = {x1, x2, x3, ..., x k}, then determine the corresponding k categories based on the secondary mapping of the AI ​​model, and then determine the corresponding k configuration sets P = {p1, p2, p3, ..., p k Then, the target network configuration information is determined according to the k configuration sets P. For example, the target network configuration information can be determined by taking an average value or a weighted average value of the k configuration sets P.

[0114] In the embodiment of the present application, the network device can calculate the nearest neighbor distance D based on the Euclidean distance Dis. m , you can also calculate the nearest neighbor distance D based on other distance calculation methods m For example, the nearest neighbor distance D can be calculated based on the Manhattan distance calculation method. m , no specific limitation is made here.

[0115] In an embodiment of the present application, the network device can determine the most matching model sample based on distance, that is, the nearest model sample. In addition, the most matching model sample can also be determined according to other methods, such as determining the similarity between the target scene feature information and the model sample, etc., which are not limited here.

[0116] 705. Network equipment determines the reliability of the AI ​​model.

[0117] Network devices determine the reliability of the AI ​​model in the current scenario.

[0118] In one possible implementation, the network device can determine the reliability of the AI ​​model in the current scenario through the nearest neighbor distance of the model, and the reliability of the AI ​​model indicates the accuracy of the target network configuration information determined based on the AI ​​model.

[0119] In one possible implementation, when determining the nearest neighbor distance D m After that, if the nearest neighbor distance D m If the reliability threshold D is smaller than the reliability threshold, the reliability P of the AI ​​model can be determined. m =1, that is, the AI ​​model is reliable; if the nearest neighbor distance D m If it is greater than the reliability threshold D, the reliability P of the AI ​​model can be determined. m =0, which means that the AI ​​model is unreliable.

[0120] In another possible implementation, when determining the nearest neighbor distance D m After that, we can use the nearest neighbor distance D m Calculate the reliability of the AI ​​model. The reliability can be a value between 0 and 1. The larger the value, the better the reliability of the AI ​​model. For example, the reliability of the AI ​​model P m ∝α / D m , P m With D m The value is inversely proportional to the value of P. m The value of the parameter is ultimately a value between 0 and 1.

[0121] In an embodiment of the present application, the network device can determine the reliability of the AI ​​model in the current scenario through the nearest neighbor distance of the model. In addition, the reliability of the AI ​​model in the current scenario can also be determined by other means, such as calculating whether the time since the last optimization and adjustment of the AI ​​model exceeds a threshold. If the threshold is exceeded, it is determined that the AI ​​model is unreliable.

[0122] 706. When it is determined that the AI ​​model is reliable, the network device sends the target network configuration information.

[0123] When the network device determines that the AI ​​model is reliable based on the reliability, the network device sends the target network configuration information to the network environment.

[0124] Specifically, the network device can send the target network configuration information to the corresponding device. For example, if the AI ​​model is used to optimize the traffic-related configuration of switch A, the network device can send the target network configuration information to switch A.

[0125] In the embodiment of the present application, through steps 701 to 706, the network device sends the target network configuration information when determining that the AI ​​model is reliable, thereby ensuring the reliability of the sent network configuration information and being suitable for the current network environment.

[0126] In a possible implementation, the network optimization method provided in the embodiment of the present application further includes step 707, which is specifically as follows:

[0127] 707. The network device determines an aging parameter of a model sample of the AI ​​model.

[0128] The network device determines an aging parameter of the AI ​​model, where the aging parameter indicates an aging degree of the AI ​​model.

[0129] In one possible implementation, the network device may determine the total number of hits of the model samples of the AI ​​model and the number of hits within a period T, and then determine the aging index of the AI ​​model based on the total number of hits and the number of hits within the period T.

[0130] Specifically, as shown in Table 1, the network device can record the total number of hits of the model samples of the AI ​​model and the number of hits within a period T.

[0131] Model Item Number of hits / S <![CDATA[Number of hits within period T / S t > A 10 8 B 2 0 … … … M 5 4

[0132] Table 1

[0133] Then, the aging index of the AI ​​model is calculated based on the information in Table 1. For example, the score function P of the model item can be used. i =α*S+β*S t To calculate the aging index of the model item. For example, α is 1 and β is 2, then the score function P of model item A is A =α*S+β*S t =1*10+2*8=26; score function P of model item B B =α*S+β*S t =1*2+2*0=2; score function P of model term M M =α*S+β*S t =1*5+2*4=13.

[0134] After the network device determines the score function of each model item, if the score function is lower than the threshold, it is determined that the model item has aged. For example, the threshold L of the score function a is 10, the score function P of model term A A = 26, model item A is not aged; the score function of model item B is P B =2, model item B has aged.

[0135] 708. Network devices delete aging model samples of AI models.

[0136] When the aging index of the model item does not meet the requirements, the aged model samples are deleted.

[0137] For example, the aging index of model item B does not meet the requirements, model item B has aged, and the network device deletes the aged model sample of the AI ​​model, namely, model item B.

[0138] In the embodiment of the present application, step 707 and step 708 are checks on the model items of the AI ​​model, so steps 707 and 708 are not necessarily after step 706, and can be triggered and executed at any time in the embodiment of the present application, and are not specifically limited here.

[0139] 709. The network device determines whether the AI ​​model needs error correction.

[0140] The network device determines whether the AI ​​model needs error correction.

[0141] Specifically, after the network device sends the target network configuration information, the network device needs to detect the network environment in which the target network configuration information is applied, so as to determine the application effect of the target network configuration information. The network device can determine whether the AI ​​model of the network device needs error correction based on feedback information such as network feature information of the network environment.

[0142] In a possible implementation, the network device obtains network characteristic change information based on the network environment, such as the network bandwidth change information ΔM=(M1, M2, M3, ..., M N ), N is the information dimension of the network bandwidth change information. Then the network device defines the revenue function RWD = α*M1+β*M2+γ*M3+…+θ*M N , through the profit function RWD, the corresponding profit of the network environment after applying the target network configuration information can be determined. If the profit of the network drops a lot after applying the target network configuration information, it means that the target network configuration information is not optimal, and the AI ​​model needs to be corrected.

[0143] When determining whether to perform error correction, the network device may define an error correction function E to perform model error correction. The error correction function E is as follows:

[0144]

[0145] Among them, C(RWD) is the model opportunity function, f t (RWD) is the RWD yield curve in period t,

[0146]

[0147] f t (RWD) = a*RWD t +b*RWD t-1 +…+k*RWD1.

[0148] When the error correction function E=0, it means that the AI ​​model does not need error correction; when the error correction function E=1, it means that the AI ​​model needs error correction.

[0149] In a possible implementation, the network device may not use the model opportunity function C(RWD) for error correction. The error correction function E is as follows:

[0150]

[0151] f t (RWD) = a*RWD t +b*RWD t-1 +…+k*RWD1.

[0152] The network device can determine whether the AI ​​model needs error correction based only on the benefit function RWD. When the error correction function E=0, it means that the AI ​​model does not need error correction; when the error correction function E=1, it means that the AI ​​model needs error correction.

[0153] In an embodiment of the present application, the network device can limit the calculation of the error correction function within period t. In addition, it can also be limited to other time periods, such as the most recent period, or multiple periods, and the corresponding weights can be set according to the distance of the multiple periods from the current time and the duration of the multiple periods. No specific limitation is made here.

[0154] In one possible implementation, as shown in step 704, the network device can determine the first k samples with the closest neighbor distance, and then determine the corresponding k configuration sets based on the secondary mapping of the AI ​​model, and then determine the target network configuration information according to the k configuration sets P. In this possible implementation, the network device can also define an error correction function E based on the k configuration sets, and the configuration set with a closer neighbor distance has a larger corresponding weight. For example, k configuration set P = {p1, p2, p3, ..., p k}, error correction function E = a*p1+b*p2+c*p3+…+d*p k , p1 to pk The neighbor distance increases successively, and a, b, c, ..., d are the corresponding weight parameters that decrease successively. When the error correction function E is greater than the threshold, the network device can determine that the AI ​​model needs to be corrected. In this case, the network device can perform error correction operations on one or more model samples.

[0155] In an embodiment of the present application, the network device may correct the AI ​​model by deleting the corresponding AI model sample. In addition, other methods may be used, such as suspending the use of the AI ​​model sample, etc., which are not limited here.

[0156] In an embodiment of the present application, through steps 707 to 709, the network device can perform aging detection and error correction detection on the AI ​​model, and perform error correction and aging processing on the AI ​​model, thereby realizing self-update and real-time error correction of the AI ​​model and enhancing the reliability of the AI ​​model.

[0157] In one possible implementation, after the network device determines the reliability of the AI ​​model in step 705, the network device determines that the AI ​​model is unreliable based on the reliability. In this possible implementation, the network optimization method provided in the embodiment of the present application further includes:

[0158] 710. When the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0159] like Figure 8 As shown, when the network device determines that the AI ​​model is unreliable based on the reliability, the target network configuration information can be determined through configuration exploration. The specific configuration exploration steps are as follows:

[0160] S1. The network device obtains network status information.

[0161] The network device periodically obtains network status information of the network environment, such as network traffic change information.

[0162] S2. The network device determines that the network state is stable according to the network state information.

[0163] After the network device obtains the network status information, it determines that the state of the network environment is stable according to the network status information, and then the network device can perform subsequent network configuration exploration steps.

[0164] In a possible implementation, if the network environment state is unstable and the explored network configuration information is not applicable to the subsequent network, the subsequent steps are temporarily not executed until the network environment state becomes stable.

[0165] In a possible implementation, the network device may not need to detect whether the network status is stable and may directly perform configuration exploration, that is, it may not execute step S1 and step S2 and directly start executing subsequent steps, which is not specifically limited here.

[0166] S3. The network device performs feedback network configuration exploration.

[0167] The network device performs feedback network configuration exploration to explore the target network configuration information.

[0168] In the embodiment of the present application, when the network device performs feedback network configuration exploration, a relevant algorithm function may be defined, as follows:

[0169] Define the feedback function: Among them, F i is the change of the i-th dimension network feature, μ i is the weight of the feature change of the i-th dimension, and ρ is the coefficient. The feedback function RWD reflects the network feedback based on a certain exploration action. The network device can determine the next exploration direction based on the feedback of the network.

[0170] Define the profit function: PROF = f(RWD). The profit function is a profit curve based on the feedback function. Based on the profit curve, we can determine whether the exploration has converged.

[0171] Define the optimization objective function: TGT = (T1, T2, T3, ..., T K ), the optimization objective function is the optimization objective defined by the network device based on the network characteristics, and the network device determines whether to stop the network configuration exploration based on the achievement of the optimization objective.

[0172] Define optimization event: EVE = (E α , E β , …, E γ ), explore the event sorting based on the weights of achieving the objective function based on different optimization events, and construct action execution units in sequence.

[0173] Define exploration probability: P e ∝Dis(CUR, TGT), according to the distance between the current network feature CUR and the target function TGT, the random exploration probability is calculated, and the network device determines whether to perform random exploration.

[0174] In the embodiment of the present application, the network device can define relevant algorithm functions. In addition, other algorithm functions and other information can also be defined, which are not specifically limited here. It can be understood that these algorithm functions defined by the network device in the embodiment of the present application can not only be used in step S3, but also in subsequent steps S4-S6, as described in the subsequent description for details.

[0175] Specifically, Fig. 9 As shown, when the action execution unit of the network device performs feedback network configuration exploration, when exploring in direction 1, if the feedback function RWD value increases, then continue to explore in direction 1; if the feedback function RWD value decreases, then retreat to direction 1 and explore in direction 2. If the feedback function RWD value does not increase or decrease when exploring in direction 1, then exploration in direction 2 can be performed.

[0176] S4. The network device determines that the configuration exploration has converged.

[0177] The network device determines that the configuration discovery has converged.

[0178] Specifically, the network device may determine that the configuration exploration has converged according to the defined profit function PROF.

[0179] In a possible implementation, if the network device determines that the configuration exploration has not converged according to the profit function PROF, the network device continues to perform the configuration exploration.

[0180] In the embodiment of the present application, when the configuration exploration of the network device has converged but has not reached the optimization target, the network device may perform random exploration in step S5.

[0181] S5. Network devices perform random exploration.

[0182] When the configuration exploration of the network device has converged but has not reached the optimization goal, for example, when the configuration exploration of the network device falls into a local optimum, the network device can perform random exploration.

[0183] Specifically, for example, the network device calculates the random exploration probability according to the distance between the current network feature CUR and the target function TGT, and the network device determines that random exploration is needed, and the network device can perform random exploration.

[0184] S6. The network device determines the target network configuration information.

[0185] After performing configuration exploration and / or random exploration, the network device determines target network configuration information according to the exploration results.

[0186] like Fig.10 As shown, in the embodiment of the present application, when the network device performs configuration exploration, the exploration algorithm is based on the event transition between states, and each action state is composed of multiple action execution units. Fig.10 As shown:

[0187] Initial state: the initial state of the state machine, starting to monitor network traffic, thereby obtaining network traffic change information, corresponding to step S1;

[0188] Detection state: periodically monitor network traffic changes, determine whether the traffic is stable, and decide whether to enter configuration exploration, corresponding to step S2;

[0189] Action state: the execution module of the exploration algorithm, which performs feedback exploration based on the action execution unit, corresponding to step S3;

[0190] Convergence state: The exploration algorithm enters the convergence state and no longer conducts exploration, but continues to monitor network changes, corresponding to step S4;

[0191] Exploration state: Based on the distance to the exploration target, exploration is performed randomly according to probability, corresponding to step S5;

[0192] Target state: The exploration algorithm achieves the exploration goal and continuously monitors network changes, corresponding to step S6.

[0193] 711. The network device determines a target sample category corresponding to the target network configuration information determined through configuration exploration.

[0194] After the network device determines the target network configuration information through configuration exploration, the network device determines the target sample category corresponding to the target network configuration information.

[0195] Specifically, the network device determines the similarity P between the target network configuration information and the sample category of the AI ​​model c .

[0196]

[0197] Among them, x i To explore the i-th dimension value of the algorithm output item x, y i The closest distance D m The corresponding parameter item has the value of the i-th dimension. The parameter items have n dimensions in total. Based on the Manhattan distance, the similarity P can be calculated. c Then according to the similarity P c Determine the corresponding target sample category.

[0198] In a possible implementation, the corresponding target sample category does not exist in the AI ​​model of the network device. Therefore, the network device can add a new target sample category in the AI ​​model, and the target sample category corresponds to the target network configuration information.

[0199] 712. The network device updates the sample set information of the AI ​​model.

[0200] After determining the target network configuration information and the target sample category through configuration exploration, the network device can update the sample set information of the AI ​​model together with the target scene feature information determined in step 703. For example, a model sample can be added to the AI ​​model, and the model sample includes the target network configuration information, the target sample category and the target scene feature information.

[0201] In the embodiment of the present application, through steps 710 to 712, the network device can determine the optimal configuration through online configuration exploration, and can also conduct random exploration based on the distance between the convergence status of the current exploration and the target state to prevent local optimality. This enables autonomous real-time online incremental learning of the AI ​​model without manual intervention, ensures good generalization of the model in unknown scenarios, and ensures the optimization effect of unknown scenarios.

[0202] The embodiment of the present application provides a network device 1100. In the embodiment of the present application, the network device 1100 can be divided into functional modules according to the above method example. For example, each functional module can be divided corresponding to each function, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present invention is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0203] In the case of dividing each functional module into corresponding functional modules, Fig.11 A possible structural diagram of the network device 1100 involved in the above embodiment is shown. Fig.11 As shown, the network device 1100 includes:

[0204] The construction module 1101 is used to construct an AI model according to offline network data. For example, the step 701 in the above method embodiment is performed: the network device constructs an AI model.

[0205] The construction module 1101 is specifically used to determine the sample set information of the AI ​​model based on the offline network data, and the sample set information includes scene feature information, category identification and network configuration information, wherein the scene feature information includes the bandwidth, delay, number of packet losses and congestion status of the network data, the category label indicates the sample category of the scene feature information, and the optimal configuration set indicates the network configuration information corresponding to the sample category of the scene feature information. For example, step 701 in the above method embodiment is executed: the network device constructs the AI ​​model.

[0206] The acquisition module 1102 is used to acquire target network data; for example, the step 702 in the above method embodiment is executed: the network device acquires the target network traffic data.

[0207] The first determination module 1103 is used to determine the target scene feature information based on the artificial intelligence AI model according to the target network data; for example, execute step 703 in the above method embodiment: the network device determines the target scene feature information based on the AI ​​model.

[0208] The second determination module 1104 is used to determine the target network configuration information according to the target scene feature information; for example, executing step 704 in the above method embodiment: the network device determines the target network configuration information according to the target scene feature information.

[0209] In a possible implementation, the second determining module 1104 includes:

[0210] The first determination unit 1105 is used to determine the target sample set information in the corresponding AI model according to the target sample feature information; for example, executing step 704 in the above method embodiment: the network device determines the target network configuration information according to the target scene feature information.

[0211] In a possible implementation, the first determining unit 1105 includes:

[0212] The first determination subunit 1106 is used to determine the nearest neighbor distance between the target sample feature information and the AI ​​model, where the nearest neighbor distance is the nearest neighbor distance between the target sample feature information and multiple sample set information of the AI ​​model; for example, executing step 704 in the above method embodiment: the network device determines the target network configuration information based on the target scene feature information.

[0213] The second determination subunit 1107 is used to determine the target sample set information of the AI ​​model corresponding to the nearest neighbor distance. For example, step 704 in the above method embodiment is performed: the network device determines the target network configuration information according to the target scene feature information.

[0214] In another possible implementation, the first determining unit 1105 includes:

[0215] The third determination subunit 1108 is used to determine k sample set information of the AI ​​model that is adjacent to the target sample feature information; for example, executing step 704 in the above method embodiment: the network device determines the target network configuration information according to the target scene feature information.

[0216] The fourth determining subunit 1109 is used to determine k network configuration information corresponding to the k sample set information; for example, executing step 704 in the above method embodiment: the network device determines the target network configuration information according to the target scene feature information.

[0217] The fifth determining subunit 1110 is configured to determine target network configuration information according to k network configuration information. For example, step 704 in the above method embodiment is performed: the network device determines the target network configuration information according to the target scenario feature information.

[0218] The second determining unit 1111 is used to determine the corresponding target category identifier according to the target sample set information; for example, execute step 704 in the above method embodiment: the network device determines the target network configuration information according to the target scene feature information.

[0219] The third determining unit 1112 is configured to determine the corresponding target network configuration information according to the target category identifier. For example, step 704 in the above method embodiment is performed: the network device determines the target network configuration information according to the target scene feature information.

[0220] The third determination module 1113 is used to determine the reliability of the AI ​​model; for example, executing step 705 in the above method embodiment: the network device determines the reliability of the AI ​​model.

[0221] In a possible implementation, the third determining module 1113 includes:

[0222] The fourth determination unit 1114 is used to determine the nearest neighbor distance between the target sample feature information and the AI ​​model; for example, executing step 705 in the above method embodiment: the network device determines the reliability of the AI ​​model.

[0223] The fifth determining unit 1115 is configured to determine the reliability of the AI ​​model according to the nearest neighbor distance. For example, step 705 in the above method embodiment is performed: the network device determines the reliability of the AI ​​model.

[0224] The sending module 1116 is used to send the target network configuration information when the AI ​​model is determined to be reliable according to the reliability. For example, step 706 in the above method embodiment is performed: when the AI ​​model is determined to be reliable, the network device sends the target network configuration information.

[0225] In a possible implementation, the network device 1100 further includes:

[0226] The fourth determination module 1117 is used to determine the aging index of the model sample of the AI ​​model. For example, step 707 in the above method embodiment is performed: the network device determines the aging parameter of the model sample of the AI ​​model.

[0227] In a possible implementation, the fourth determining module 1117 includes:

[0228] The sixth determination unit 1118 is used to determine the total hit count of the model samples of the AI ​​model and the hit count within period T; for example, executing step 707 in the above method embodiment: the network device determines the aging parameters of the model samples of the AI ​​model.

[0229] The seventh determining unit 1119 is used to determine the aging index of the model sample of the AI ​​model according to the total hit count and the hit count within the period T. For example, step 707 in the above method embodiment is performed: the network device determines the aging parameter of the model sample of the AI ​​model.

[0230] The first deletion module 1120 is used to delete the aged model samples of the AI ​​model when the aging index does not meet the aging degree requirement of the AI ​​model. For example, step 708 in the above method embodiment is executed: the network device deletes the aged model samples of the AI ​​model.

[0231] In a possible implementation, the network device 1100 further includes:

[0232] The fifth determination module 1121 is used to determine whether the AI ​​model needs error correction; for example, executing step 709 in the above method embodiment: the network device determines whether the AI ​​model needs error correction.

[0233] Specifically, the fifth determining module 1121 includes:

[0234] The acquisition unit 1122 is used to obtain network feature change information; for example, execute step 709 in the above method embodiment: the network device determines whether the AI ​​model needs error correction.

[0235] The eighth determination unit 1123 is used to determine the error correction index according to the network characteristic change information; for example, execute step 709 in the above method embodiment: the network device determines whether the AI ​​model needs error correction.

[0236] The ninth determining unit 1124 is configured to determine whether the AI ​​model needs error correction according to the error correction index. For example, step 709 in the above method embodiment is performed: the network device determines whether the AI ​​model needs error correction.

[0237] The second deletion module 1125 is used to delete the model samples that need error correction in the AI ​​model when the AI ​​model needs error correction. For example, step 709 in the above method embodiment is performed: the network device determines whether the AI ​​model needs error correction.

[0238] In a possible implementation, the network device 1100 further includes:

[0239] The sixth determination module 1126 is used to determine the target network configuration information through configuration exploration when the AI ​​model is determined to be unreliable according to the reliability. For example, step 711 in the above method embodiment is performed: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0240] Specifically, the sixth determination module 1126 includes:

[0241] The acquisition unit 1127 is used to obtain network status information; for example, execute step 711 in the above method embodiment: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0242] The tenth determination unit 1128 is used to determine whether the network state is stable according to the network state information; for example, execute step 711 in the above method embodiment: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0243] The exploration unit 1129 is used to perform feedback network configuration exploration when the network state is determined to be stable. For example, step 711 in the above method embodiment is performed: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0244] The eleventh determination unit 1130 is used to determine whether the configuration exploration has converged; for example, executing step 711 in the above method embodiment: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0245] The random exploration unit 1131 is used to perform random exploration. For example, step 711 in the above method embodiment is performed: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0246] The twelfth determining unit 1132 determines the target network configuration information. For example, step 711 in the above method embodiment is performed: when the AI ​​model is unreliable, the network device determines the target network configuration information through configuration exploration.

[0247] The seventh determination module 1133 is used to determine the corresponding target sample category according to the target network configuration information; for example, executing step 711 in the above method embodiment: the network device determines the target sample category corresponding to the target network configuration information determined by configuration exploration.

[0248] The new module 1134 is used to add a target sample category corresponding to the target network configuration information in the AI ​​model. For example, step 711 in the above method embodiment is performed: the network device determines the target sample category corresponding to the target network configuration information determined by configuration exploration.

[0249] The updating module 1135 is used to update the sample set information of the AI ​​model according to the target network configuration information, the target sample category and the target scene feature information. For example, the step 712 in the above method embodiment is executed: the network device updates the sample set information of the AI ​​model.

[0250] The various modules of the above-mentioned network device can also be used to perform other actions in the above-mentioned method embodiment. All relevant contents of each step involved in the above-mentioned method embodiment can be referred to the functional description of the corresponding functional module and will not be repeated here.

[0251] Fig.12 It is a schematic diagram of a network device structure provided in an embodiment of the present application. The network device 1200 may include one or more central processing units (CPU) 1201 and a memory 1205. The memory 1205 stores one or more application programs or data.

[0252] The memory 1205 may be a volatile storage or a persistent storage. The program stored in the memory 1205 may include one or more modules, each of which may include a series of instruction operations in the network device. Furthermore, the central processor 1201 may be configured to communicate with the memory 1205 and execute a series of instruction operations in the memory 1205 on the network device 1200.

[0253] The central processor 1201 is used to execute the computer program in the memory 1205, so that the network device 1200 is used to perform: obtaining target network data; determining target scene feature information based on the artificial intelligence AI model according to the target network data; determining target network configuration information according to the target scene feature information; determining the reliability of the AI ​​model; and sending the target network configuration information when the AI ​​model is determined to be reliable according to the reliability. For specific implementation methods, please refer to Figure 7 Steps 701 - 712 in the illustrated embodiment will not be described in detail here.

[0254] The network device 1200 may also include one or more power supplies 1202, one or more wired or wireless network interfaces 1203, one or more input and output interfaces 1204, and / or one or more operating systems, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0255] The network device 1200 can perform the aforementioned Fig.12 The operations performed by the network device in the illustrated embodiment will not be described in detail here.

[0256] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using a software program, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, all or part of the processes (or functions) described in the embodiments of the present application are implemented. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or may contain one or more servers, data centers and other data storage devices that can be integrated with the medium. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium (eg, a solid state disk (SSD)), etc. In the embodiment of the present application, the computer may include the aforementioned device.

[0257] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in a claim. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

Claims

1. A network optimization method, characterized in that: The method comprises: Get target network data; Determine target scene feature information based on an artificial intelligence AI model according to the target network data; Determine target network configuration information according to the target scene feature information; Determining the reliability of the AI ​​model; When it is determined that the AI ​​model is reliable according to the reliability, the target network configuration information is issued.

2. The method according to claim 1, characterized in that The method further comprises: The AI ​​model is constructed based on offline network data.

3. The method according to claim 2, characterized in that The AI ​​dictionary model construction includes: The sample set information of the AI ​​model is determined according to the offline network data, and the sample set information includes scene feature information, category identification and network configuration information, wherein the scene feature information includes the bandwidth, latency, number of packet losses and congestion status of the network data, the category label indicates the sample category of the scene feature information, and the optimal configuration set indicates the network configuration information corresponding to the sample category of the scene feature information.

4. The method according to any one of claims 1 to 3, characterized in that: The determining the target network configuration information according to the target scene feature information includes: Determine the target sample set information in the corresponding AI model according to the target sample feature information; Determine a corresponding target category identifier according to the target sample set information; The corresponding target network configuration information is determined according to the target category identifier.

5. The method according to claim 4, characterized in that The determining the target sample set information in the corresponding AI model according to the target sample feature information includes: Determine the nearest neighbor distance between the target sample feature information and the AI ​​model, where the nearest neighbor distance is the nearest neighbor distance between the target sample feature information and multiple sample set information of the AI ​​model; Determine the target sample set information of the AI ​​model corresponding to the nearest neighbor distance.

6. The method according to claim 4, characterized in that The determining the target network configuration information according to the sample characteristic information includes: Determine k sample set information of the AI ​​model that is adjacent to the target sample feature information; Determine k pieces of network configuration information corresponding to the k pieces of sample set information; The target network configuration information is determined according to the k network configuration information.

7. The method according to claim 5 or 6, characterized in that: Determining the reliability of the AI ​​model includes: Determine the nearest neighbor distance between the target sample feature information and the AI ​​model; The reliability of the AI ​​model is determined based on the nearest neighbor distance.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Determine an aging index of a model sample of the AI ​​model.

9. The method according to claim 8, characterized in that The determining of the aging parameters of the model samples of the AI ​​model includes: Determine the total number of hits of the model samples of the AI ​​model and the number of hits within a period T; The aging index of the model sample of the AI ​​model is determined according to the total hit count and the hit count within the period T.

10. The method according to claim 9, characterized in that The method further comprises: When the aging index does not meet the aging degree requirement of the AI ​​model, the aged model samples of the AI ​​model are deleted.

11. The method according to any one of claims 1 to 10, characterized in that: The method further comprises: Determining whether the AI ​​model needs error correction; When the AI ​​model needs error correction, the model samples that need error correction in the AI ​​model are deleted.

12. The method according to claim 11, characterized in that Determining whether the AI ​​model needs error correction includes: Obtain information about changes in network characteristics; Determining an error correction index according to the network characteristic change information; Determine whether the AI ​​model needs error correction based on the error correction index.

13. The method according to claim 1, characterized in that The method further comprises: When it is determined that the AI ​​model is unreliable according to the reliability, target network configuration information is determined through configuration exploration.

14. The method according to claim 13, characterized in that Determining the target network configuration information through configuration exploration includes: Conduct feedback network configuration exploration; Determining that the configuration exploration has converged; The target network configuration information is determined.

15. The method according to claim 14, characterized in that Before performing the feedback network configuration exploration, the method further includes: Get network status information; Determining that the network state is stable according to the network state information; The feedback network configuration exploration comprises: When the network status is determined to be stable, feedback network configuration exploration is performed.

16. The method according to claim 14 or 15, characterized in that When the target network configuration information does not meet the preset requirements, the method further includes: Conduct random exploration; Re-determine the target network configuration information.

17. The method according to claim 16, characterized in that The method further comprises: Determine a corresponding target sample category according to the target network configuration information.

18. The method according to claim 16, characterized in that The sample set information of the AI ​​model does not include the target sample category, and the method further includes: A target sample category corresponding to the target network configuration information is added to the AI ​​model.

19. The method according to claim 17 or 18, characterized in that The method further comprises: Update the sample set information of the AI ​​model according to the target network configuration information, the target sample category and the target scene feature information.

20. A network device, characterized in that: The network equipment includes: An acquisition module, used to acquire target network data; A first determination module is used to determine target scene feature information based on an artificial intelligence AI model according to the target network data; A second determination module is used to determine target network configuration information according to the target scene feature information; A third determination module, used to determine the reliability of the AI ​​model; A sending module is used to send the target network configuration information when it is determined that the AI ​​model is reliable according to the reliability.

21. A network device, characterized in that: The network device comprises a processor and a memory; the processor is coupled to the memory; the memory is used to store computer instructions, and the computer instructions are loaded and executed by the processor to enable the network device to implement the method according to any one of claims 1-19.

22. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program instruction, and the computer program instruction is loaded and executed by a processor to implement the method according to any one of claims 1 to 19.